Middle-East Journal of Scientific Research 25 (7): 1619-1623, 2017 ISSN 1990-9233 IDOSI Publications, 2017 DOI: 10.5829/idosi.mejsr.2017.1619.1623 3D Face Recognition System in Time Critical Security Applications 1 2 R. Reji and P. Sojan Lal 1 Research Scholar, School of Computer Sciences, M G University, Kottayam, Kerala, India 2 Principal, Mar-Baselious Institute of Technology and Science, Kothamangalam, Kerala, India Abstract: 3D face recognition has gained lot of attention due to improved sensors and advanced algorithms, deployment of this modality in biometrics systems is common now days. This paper presents the application of Region based 3D face recognition system. Region based face recognition system works by extracting 15 small regions from the frontal face; Modified face recognition algorithm along with hierarchical contour based registration technique is applied for finding similarity. We are operating this system in two modes namely the verification mode and the confirmation mode. The approaches employed is Distributed computing which gives more insight into the implementation of the system in time critical applications. Key words: Biometrics 3D Face Recognition 2D Face Recognition MA INTRODUCTION In 3D face recognition the use of geometric depth information is having more relevance than color and The advancement in Computer technology and the texture; it is invariant to head angles, camera distance. need for better security applications brought biometrics Thus most of the limitations of 2D face recognition can be into the main scenario. The term biometrics refers to the resolved by using 3D face recognition approaches, refer calculation of unique physical or behavioral Table 1. 3D face recognition dominates mainly due to the characteristics for verifying personal identity. Face advancement in 3D sensors. recognition is one of the most important research areas in The major application of face recognition in earlier computer vision and image processing. This technique is days was in access control and video surveillance. This the least intrusive and most popular among biometric technology now plays a vital role in access and security, modalities. Face is considered as the most attractive payments, criminal identification and health care. More biometrics due to its public acceptance. In Biometrics over face recognition algorithms are implemented in video context the word recognition can be defined as the gaming and artist s uses facial recognition technology to capability to perform identification and verification. In project digital makeup on models. verification one biometric pattern is compared with Different algorithms are proposed to address the another biometric pattern whereas in identification, one diverse aspects of face recognition technology, so here biometric pattern is compared with a set of biometric arise the need to analyze the effectiveness and efficiency patterns. of each algorithm. In this paper we are focusing on the Face recognition is considered as a difficult pattern effectiveness of region based 3D face recognition system recognition problem mainly due to inter class similarity powered by modified face recognition algorithm. and intra class variability. The intra class variability may be due to pose change, illumination, expressions, facial Related Work and Overview: A detailed survey of face accessories and aging effect. Face recognition area is recognition with its features was given in [1]. broadly classified as 2D face recognition and 3D face Gokberg et al. [2] discuss about the advancement in recognition. The major problem with 2D face recognition 3D face recognition technology, current research trends process is the change in pose, illumination and and open challenges. Five real world scenarios where 3D expression. As a result of this 2D face recognition system face recognition can be applied are highlighted in his can be employed in limited applications. work. Corresponding Author: R. Reji, Research Scholar, School of Computer Sciences, M G University, Kottayam, Kerala, India. 1619
Table 1: Strengths of 2D /3D Face recognition Features 2D Face Recognition 3D Face Recognition Accuracy Ease of use Cost Precision Acceptance Akarun et al. [3] highlights on the 3D face recognition and its biometrics applications. Chang et al. [4] makes use of a region based 3D face recognition approach by dividing the face into multiple sub regions. These regions are located in and around the nose. Faltemier et al. [5] proposed a region based approach by dividing the face region into 28 sub region They reported a Rank one recognition rate of 97.2% and VR of 93.2% at an FAR of 0.1%. They further extend their work by taking the number of regions to 38. Reji et al. [6] presented a region based approach by dividing the face region into 48 sub regions and reported a Rank one recognition rate of 97.1% and VR of 93.7% at FAR of 0.1%. Zhong et al. [7] divides the frontal face into, the upper face region and the lower face region. The upper face region without the mouth is used for experimentation. K-means clustering is applied and results were obtained using nearest neighbor classifier. Reji et al. [8] propose an algorithm for analyzing altered fingerprints along with its software implementation in java. Lie et al. [9] presents a 3D face recognition approach relying on low level geometric features that are collected from the forehead, eyes and nose. These regions are relatively unchangeable in the presence of facial expressions. Proposed System: Our region based 3D face recognition system is having two modes of operation, the Verification mode and Confirmation mode. This face recognition system is implemented as two programs using IDL as the language. The whole system is tested in a distributed environment with parallel processing. Architecture of our proposed face recognition system is shown in figure 4. We can apply this system mainly for criminal identification purpose. In our approach we are having three databases namely test database, trained and probe database. Test database is having a collection of criminal s data. Trained database is used for storing sub regions; this will act as gallery while comparing with the image under scrutiny. Probe database is temporarily storing the image under scrutiny. Implementation of this approach starts with the execution of program 1. The algorithmic approach in this program is for automatic detection of nose tip and sub region generation. Steps in program 1: Input the image. Smooth the image Automatic detection of nose tip. Sub region generation Alignment of sub region to trained database. Output obtained from program 1is highlighted in Figure 1, 2. These steps are repeated for all the face images in the test database. The next phase of this system is implemented in. The major algorithmic approach in this program is the Modified Face Recognition algorithm and the Hierarchical contour based image registration. When a new face arrives we need to check whether it is having any similarity with the images in test database. Steps in : Input the image under scrutiny. Copy it into probe database. Apply MA. Calculate the rank based similarity score. Report match or not. Move to the next mode if required. Apply Hierarchical contour based image registration Match confirmed / No Match Stop. The similarity measure is calculated and is compared against a threshold value. If the similarity score is greater than or equal to threshold a match is obtained otherwise no match, refer figure 3.At some cases the similarity score is just less than the specified threshold value. This situation adds some fuzziness to the system. In this case the second phase of MA called confirmation phase comes to play. Rank based score is taken; hierarchical contour based image registration is applied to find a match with the image in scrutiny and the most matched image from the test database. 1620
Fig. 1: Program 1- Test image Fig. 2: Program 1- Sub Region generation Fig. 3: Program 2- Matched Image 1621
Fig. 4: Architecture diagram Legends: Face Recognition DB- Data Base SERVER program 1 Test DB Trained DB CLIENT 1 CLIENT 2 CLIENT N Fig. 5: Implementation of system 1622
Table 2: Running time in Client Phase Steps Time(ms) Verification Mode Data Preprocessing 4, 540 Matching 1, 940 Confirmation Mode Registration 5, 120 Table 3: Running time in Server Phase Steps Time(ms) Verification Mode Data Preprocessing 2, 070 Matching 645 Confirmation Mode Registration 2, 150 System Level Implementation: Region based 3D face recognition frame work is implemented and tested in a Distributed computing approach. Distributed computing is conceptually closer to parallel computing. Program 1 is loaded on the server with multiple cores. The test database and the trained database are stored on the server. The is loaded on client machine. When a new image under scrutiny is obtained the client machine store the image temporarily on the probe database. The MA algorithm compares the image from the client with trained database in the server. The system level implementation is shown in Figure 5. Table 2, 3 shows the running time of the region based face recognition system in verification mode and confirmation mode. In case of a client from Preprocessing to hierarchical image registration technique the process takes less than 12 Seconds on a 2.40 GHz Intel Core i3 Processor with 4 GB of memory. This suggests that a feasible execution time may be achieved for use at critical security applications. CONCLUSION The region based 3D face recognition approach is applied on Bosphorus 3d datasets and achieved a VR of 95.3% at FAR of 0.1%. In the identification scenario, rank one recognition rate of 99.3% is achieved. We are now experimenting the approach in a third party dataset also. Our system can be implemented in time critical areas such as Airport checkpoints, ATM and can be used in building other security scenario. We can speed up the face recognition system further by optimizing the IDL code. REFERENCES 1. Zhao, W., R. Chellappa, P.J. Phillips, A. Rosenfeld, 2003. Face recognition: A literature survey, ACM Comput. Surv., 35: 399-458. 2. Gokberk, Berk, Salah, Albert, Alyuz, Nese and Akarun, Lale, 2009. 3D Face Recognition: Technology &Applications. 217-246.10.1007/978-1-84882-385-3_9 3. Akarun, L., B. Gkberk and A.A. Salah, 2005. "3D face recognition for biometric applications", presented at th the Proc. 13 Eur. Signal Process. Conf. (EUSIPCO), 2005-Sep. 4. K. Chang, K., K.W. Bowyer and P. Flynn, 2006. Multiple nose region matching for 3D face recognition under varying facial expression, IEEE Trans. Pattern Anal. Mach. Intell., 28(10): 1-6. 5. Faltemier, Timothy C., Kevin W. Bowyer and Patrick J. Flynn, 2008. A Region Ensemble for 3-D Face Recognition. IEEE Transactions on Information Forensics and Security, 3(1): 62-73. doi:10.1109/tifs.2007.916287. 6. Reji, R. and S. Ravi, 2010. Comparative Analysis in 3D face Recognition, 2010-Special Issue International Journal of Imaging Science and Engineering. ISSN 1934-9955 7. Zhong, C., Z. Sun and T. Tan, 2007. Robust 3D face recognition using learned visual codebook, in: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp: 1-6. 8. Reji, R. and Akhil Mathew Philip, 2015. Altered Fingerprints: Identification and Analysis. International Journal of Computer Applications (0975 8887)International Conference on Emerging Trends in Technology and Applied Sciences (ICETTAS 2015:10-13, September 2015 9. Lei, Y., M. Bennamoun and A.A. El-Sallam, 2013. An efficient 3D face recognition approach based on the fusion of novel local low-level features. Pattern Recognition, 46(1): 2437. https://doi.org/10.1016/j.patcog.2012.06.023. 1623